Pyramidal neural networking for mammogram tumour pattern recognition

There has been much interest recently in developing neural networks to solve complicated information processing problems such as automatic diagnosis of X-ray mammograms. In New Zealand we are investigating a pyramidal neural network and adaptive contrast enhancement image processing technique for developing a knowledge-system for medical image interpretation. In this paper, we present the pyramidal network architecture with experimental breast cancer tumour pattern mapping results. The pyramidal network configuration has overcome the problem of hidden layer size. To facilitate the learning we introduce a novel method of standard coding mechanism by using local overlapping and minimum value thresholding. The outcome of this unique mapping is promising in designing a useful expert system.<>